AEC/APC Symposium 2013 looks at how to improve productivity in semiconductor processing. Big data analysis is no longer optional.
The parameters in semiconductor manufacturing are growing so large that an analysis method similar to what’s currently used for big data is now required. The good news is that big data analysis techniques, which process a vast amount of data such as search data and communication logs in the cloud, is entirely applicable to the data of semiconductor process.
Photo 1: Opening view of AEC/APC Symposium Asia 2013
There are a number of parameters used in semiconductor processes, including temperature, pressure, multiple gas flow rates and plasma power, among others, all of which are being applied in a large number of wafers. Though wafers are processed in FOUP cassettes, analyzing variation factors between casettes, inside cassettes, between wafers, and the wafer plane itself, are comparable to enormous amounts of data processing.
During the plasma process, thin film deposits on the wall of the chamber and the inner wall is shaved, so conditions may change each time a single wafer is processed. There is also a possibility that it changes between cassettes. As a result, the relationship between the process conditions and electrical characteristics of the chip is becoming very complicated.
Changing a parameter does not necessarily make the electrical properties appear as it really is. The production conditions used in semiconductor processes over the past year might not be understood physically.
Oka Akio, CIO at Toshiba Semiconductor & Storage Products Company (Photo 2), said in his keynote at the Tokyo AEC/APC Symposium Asia 2013 last month that the CIM system of factory data gathering and production equipment, and data collected daily from inspection apparatus, have reached as high as 1.6 Gigabytes of data. But compiling historical data over three years boosts that number into the petabyte range.
Photo 2: Oka Akio of Toshiba Semiconductor & Storage Products Company CIO
AEC (Advanced Equipment Control)/APC (Advanced Process Control) is the generic name of equipment and technology that control the time course and variation change of semiconductor manufacturing processing by a feed-forward approach. The conference has shown that along with the feedback/feed-forward control, statistical methods to build models has become indispensable.
Furthermore, this year, it seems that Hadoop functionality, which is significant in data center analysis, also is applicable in semiconductor process data analysis. Hadoop offers a method to parse large data across multiple computers.
The coordinated technology for producing semiconductor with high yield and uniform high-quality controls can analyze the big data. According to Mr. Oka, because it takes so long to analyze big data, there is a future need to reduce that time by opening up and standardizing the device data.
Leave a Reply